Vibration State Compressed Sensing And Fault Diagnosis Of Hydropower Units

As an important clean energy,hydropower plays an indispensable role in peak regulation and frequency regulation in our country’s power dispatch.Hydropower units are important mechanical equipment.Actively carrying out the diagnosis and maintenance work of hydropower units is conducive to timely grasping of fault information,reducing the risk of unit failures,and improving the security level of the power grid.In addition,the current development of large-scale hydropower equipment is gradually increasing speed,scale,and distribution,which has brought considerable pressure to the fault diagnosis and data processing of power stations.Therefore,this article is based on the theory of compressed sensing,focusing on the important and difficult points of hydropowerunit vibration signal processing,feature extraction,compression reconstruction,and fault diagnosis,using parameter optimization,deep learning,and sparse representation as tools,and proposes a combination of power station resources Theory application model.Under the premise of reducing sampling points,the abnormal state of the unit is accurately diagnosed.The learning dictionary is used to replace the original fixed dictionary,which effectively improves the signal sparsity and reconstruction accuracy.Based on the sparse representation theory,a deep compression reconstruction and fault diagnosis network is constructed.The main research contents and innovations of this article are as follows:(1)In order to alleviate the transmission pressure of vibration signals,the theoretical basis of compressive sensing and sparse representation is explained based on the combination of hydropower station monitoring status and compressed sensing theory.The basis of fault diagnosis and reconstruction principle of compressed signal are studied.An application model based on the safety zone of the power station is proposed to lay the foundation for the following article.(2)In order to realize the fault diagnosis based on the compressed measurement signal,a fault diagnosis model based on the compressed signal is established.In order to solve the contradiction between diagnostic performance and transmission efficiency,the compressed dimensionality interval is derived,and the concept of dimensionality decline rate is defined.An algorithm for searching the optimal compression dimension is proposed to guide the power station to determine the optimal compression dimension.In the bearing vibration signal from the vibration data set and the station disclosed operating each link is achieved,which verifies the feasibility of the compressed signal fault diagnosis,and illustrates the effectiveness and rapidity of the dimensional search algorithm.(3)The working conditions of hydropowerunits have changed complexly.In order to analyze the signal reconstruction process and obtain an adaptive sparse representation of the vibration signal,a compression reconstruction model based on dictionary learning is constructed based on sparse decomposition and dictionary learning theory.First of all,a complete dictionary,a complete dictionary,and a training dictionary are built.The influence of iteration parameters on the reconstruction process is analyzed.The decomposition sparsity and decomposition error of each dictionary are studied.The superiority of the adaptive dictionary at the sparse representation level is verified.Secondly,wavelet decomposition is used to reduce the noise of samples and improve the learning ability of the dictionary for noisy samples.Finally,in the compression and reconstruction of vibration signals,it is proved that the K-SVD dictionary has the advantages of reducing the number of iterations and improving the accuracy of reconstruction.Furthermore,in the main working condition signals,the ability of model reconstruction is verified.(4)In order to improve the combination of unit diagnosis and compressed sensing,mining data two-dimensional depth features,starting from sparse representation,a reconstruction and diagnosis network based on DCT sparse image and CNN is proposed.First,the sparseness factor is used as the index to compare the image sparsity,and the DCT dictionary is selected as the sparsity base,and the sparsity coefficient and corresponding position of the vibration signal are converted into the stripe depth and position of the sparse image to form a feature image.Secondly,with the help of the theory of block compressed sensing,from the local end to compress the sensor feature image to the analysis end,the deep convolution network is used to fit the reconstruction process,restore the feature image,and maintain the original difference of the feature.Finally,combined with the powerful feature extraction ability of CNN,the deep diagnosis network is designed to extract various types of signal features for effective state recognition...